云计算环境下虚拟机需求预测的研究控制工程专业论文.docxVIP

AbstractAbstract Abstract Abstract VM provisioning is one of the most challenging problems in cloud computing．The pay--as--you··go style of cloud computing business model makes the cloud resource allocation always fail to meet the customer’S need．Under the restrictions of many technical barriers，such as cloud resource allocation，virtual machine migration，the most practical and effective way is to predict the demands and prepare the VMs in advance． Virtual machine prediction，as a more and more important part in enterprise，it plays an increasingly significant role in cloud capacity planning decisions．The main solution is to use time series model to predict the demand of virtual machine。However,it is difficult to predict the unstable virtual machine requirements．This paper mainly focuses on two aspects of VM prediction，the suddenly changed points and the quantity of VMs．We solve these two kinds of problems by time series prediction．Specifically,Our contributions Can be summarized as follows： This paper proposes a PLR-WSVM model for predicting the sharply changed demand of VMs．In this model，we transform the prediction of VM resources into a classification issue．PLR is used to create the training samples from the original VM demand time series，and then WSVM is used to predict sharply changed demand．Our proposed method Can capture the sharply changed points in the highly unstable source demand time series and improves the prediction performance while reducing the provisioning costs．Experimental evaluation on the Smart Cloud Enterprise(SCE)trace data demonstrates the effectiveness of our proposed model． This paper proposes a RMBSE model to predict the quantity of VM demand．Firstly, all the prediction models are ranked as their past prediction accuracies at each time point， and then we get a ranking matrix of component models in a time interval．Finally we selectively combine some component models based on the ranking matrix．In addition， according to the actual application s